Stories
Can Governments Control Dangerous AI? The Debate Over Kill Switches and Open-Weight Models
N
News18•09-10-2026, 15:08
Can Governments Control Dangerous AI? The Debate Over Kill Switches and Open-Weight Models
- •China is actively integrating artificial intelligence across factories, logistics, and economic sectors using its AI Plus strategy.
- •The proliferation of open-weight AI models makes safety challenging because developers download and run models independently without centralized control.
- •The proposed US AI Kill Switch Act aims to give the Department of Homeland Security authority to suspend advanced AI systems during serious threats.
- •Distributed models like China's GLM-5.3 highlight the dual-use risks of cybersecurity capabilities when weights are publicly released.
- •Experts emphasize that AI safety cannot rely solely on centralized kill switches and must include built-in safeguards and international cooperation.
Key Questions Answered
The proposed AI Kill Switch Act would require developers of the most powerful AI systems to maintain technical shutdown mechanisms, granting the Department of Homeland Security authority to throttle or shut down models in loss-of-control scenarios.
Once model weights are released and downloaded across millions of external devices and servers, original developers lose practical control, as existing copies can continue operating offline and be modified without central oversight.
Developed by Z.ai, GLM-5.3 was assessed by the US National Institute of Standards and Technology as the most cyber-capable open-weight model released at that time, highlighting the dual-use risk that its cybersecurity features could be misused for attacks.
